Measure biggest contributors to state witness size

[Longarithm]

1. Measure which trie keys are the biggest contributors to the state witness size.

Basically we need metric (trie key first byte, shard id) -> u64. The easy way to measure it is to add (2 * key.len() + value.len()) at each Trie::get (2 because of nibbles). Alternatively we could add how much bytes each new key adds to the current state witness size, but the first way already looks like a good approximation. This can help to see why we often hit state witness size limit: https://nearone.grafana.net/goto/FaAFtMUIg?orgId=1

In the good case, if this is because of contract codes, we should do #11099 after that. Otherwise we should look for other ways to improve throughput.

2. Add metric measuring how many bytes is contributed by state items (2 * key.len() + value.len()) vs. inner trie items (everything else - basically, sizes of all CryptoHashes). This gives the merkle trie overhead.

Zulip